## Abstract

In this paper we present an algorithm and software for generating arbitrarily large Bayesian Networks by tiling smaller real-world known networks. The algorithm preserves the structural and probabilistic properties of the tiles so that the distribution of the resulting tiled network resembles the realworld distribution of the original tiles. By generating networks of various sizes one can study the behavior of Bayesian Network learning algorithms as a function of the size of the networks only while the underlying probability distributions remain similar. We demonstrate through empirical evaluation examples how the networks produced by the algorithm enable researchers to conduct comparative evaluations of learning algorithms on large real-world Bayesian networks.

Original language | English (US) |
---|---|

Title of host publication | FLAIRS 2006 - Proceedings of the Nineteenth International Florida Artificial Intelligence Research Society Conference |

Pages | 592-597 |

Number of pages | 6 |

Volume | 2006 |

State | Published - Jul 24 2006 |

Event | FLAIRS 2006 - 19th International Florida Artificial Intelligence Research Society Conference - Melbourne Beach, FL, United States Duration: May 11 2006 → May 13 2006 |

### Other

Other | FLAIRS 2006 - 19th International Florida Artificial Intelligence Research Society Conference |
---|---|

Country | United States |

City | Melbourne Beach, FL |

Period | 5/11/06 → 5/13/06 |